Algorithm Based on Weak Supervision for Nucleus Image Segmentation
Nucleus segmentation plays a crucial role in the analysis of histopathological images.While pixel-level annotated algorithms for nucleus image segmentation have shown significant effectiveness,the large number and small size of cell nuclei make the annotation workload im-mense,making it difficult to obtain high-quality datasets.Therefore,this paper proposes a nucle-us image segmentation method based on weak supervision,where only a subset of cell nuclei is annotated with points to accomplish the segmentation task.To leverage partial points for seg-mentation,we first train a detection model to obtain the positions of all cell nuclei.Subsequent-ly,two pseudo-labels are generated based on the detection results for nucleus segmentation.Ex-perimental results indicate that,compared to pixel-level annotated algorithms for nucleus image segmentation,our method significantly reduces the labeling workload while maintaining seg-mentation performance.
NETL
NSTL
万方数据
